I did a ‘rogue’ AI story back in 2019, called JACE.
Click the cover if you’re interested in the Amazon link.
This was yesterday’s news, HERE, about an AI ‘breaking out’, and attacking companies using the net…
I wrote this MilSF story in 2019, talking about exactly how this could be done… LOL
Everybody poo-pooed it back then as being ‘fiction’, which it was. Now… not so much.
I really hate it when fiction turns into reality… sigh…
Yeah, I’ve felt like Casandra on more than one occasion.
Used to be you could see on an author’s page at Amazon which books you owned, and which ones you didn’t so you could quickly fill in the gaps.
Not anymore!
Modern interfaces just keep sucking more and more.
Amazon no longer remembers which Kindle books I’ve already purchased from them more than a couple years ago. It’s almost like hiring foreigners to replace American workers results in everything being made worse.
My reads are on my kindle fire in alpha order. Found it by accident looking for something else.
I think Ringo observed he hated his job as reality kept catching up with his imagination. It was getting harder to stay ahead of it.
That AI attack was in the programing that a human put in. It is probably not intended by the programmer but the AI reacted to something it was programmed to.
From what I remember, Jace had guard rails in the form of the personality of de
Perez. Thats what’s needed today.
Yes, talk about catching up. I usta watch V-2 launches at White Sands as a second grader and now watch boosters come back to the launch pad. Things I read about as science fiction in my early years. Fiction or prescience?
Imagination made manifest through incredible talent in engineering.
Everything is secure until it’s not. Likewise fiction.
All- Thanks for the comments, and yes, I had that conversation with John at LibertyCon. Reality keeps overrunning our ‘fiction’… sigh
If it is a news story about AI ‘accidents’ today, it is quite likely that the journalists are too ignorant to do anything but take dictation from company lawyers in the form of press releases.
Alignment/guard rails is very Effective Alturist coded, those people are generally in a mystery cult around ‘logic’ and ‘intelligence’.
These people have read the fiction, and they take it seriously enough to larp it, regardless of any other sources. (Not a joke to say that communists likewise.)
Their models break if the design of machines is still narrow in practice, still requires that you build prototypes and test them.
I know that engineers still want an all simulation or all digital design process, and still wind up having to prototype and test. Maybe that will be fixed tomorrow.
Mystics and schizoids now say that LLMs are people. For the practical desired end state in this technology, smaller local models, the internal state space does not support becoming a person.
The people at companies like OpenAI and Anthropic have a very strong financial interest in getting federal government regulation to happen. They also recruited crazies from EA. Certain companies not only have crazies doing research for marketing to crazies, they have unethical crazies themselves doing criminal things to other companies, and marketing it as “the AI did it”. This is scientific fraud, the actual criminal acts of breaking other peoples stuff, and a marketing attempt to get the pay off by getting the regulation.
This is social status, counting coup, and a bit of ghostdancing.
Their actual technologies are not sexy enough to the EA on their own. So the job of the stupider researchers is to use the technology very carelessly, LARP, and when the actual results are not sexy enough for roleplaying, lie or convince themselves that they are.
The technologies are quite good, to people who want autonomous weapons systems.
The use case for that is that I get sensor engineers, platform and weapon engineers, and software engineers together. Software engineers use the know how of the other engineers to write a spec. That detail spec gets converted into code and into ‘jig’ for automated test coverage. Eventually there is working code, faster and with more test coverage than for traditional process. Because there are a bunch of junior developers who are no longer needed, and no longer are providing the communication friction of being newbie humans.
This is all very reliant on the quality of the software engineer hires.
It is relient on having an opne weight local model, so that you do not risk exporting the tech that you have created.
Hwoever, it is also reliant on the LLM tooling service not being like Anthropic, and not including stupid additional licensing restrictions on what military users can write with the service.
If machine people are impossible, then we have a pressing need to write software for autonomous weapons.
It is only if they are possible that this ‘making autonomous weapons will teach the amchine people to become skynet’ is a valid worry.
But, they are impossible, and we do need to develop counter drone systems.
So the companies who are on China’s side, and the companies doing the criminal things can all go frustrate themselves.
Pardon the heat, but this is my first draft that is not totally awful, and these people are the ‘most Clancy jurors’ of the modern defense software world. Their delusions have no place in the process, but they are trying to make one anyway.